Breath Patterns as Signals: A Machine Learning-based Molecular Communication Perspective

Exhaled breath, rich in water vapor, is a suitable medium for air-based molecular communication. This work proposes a low-cost, non-invasive framework using a DHT22 sensor to classify Eupnea, Bradypnea, and Tachypnea. Humidity and temperature signals from the mouth and nose are processed using machine learning (ML), with SHAP-based feature selection and a stacked ensemble model (XGBoost, CatBoost, Random Forest).